Autonomous Driving Peripheral And Central Vision Region Selection For Semantic Segmentation

Abstract

There has been a lot of development in autonomous cars recently and it's anticipated that fully autonomous vehicles will be on the road soon. Autonomous vehicle understanding of the surrounding environment is one of the most important autonomous driving tasks and this task is done by using deep CNN-based segmentation which imposes a remarkably high computational load. This paper proposes a novel method to reduce the computation time by mimicking the human eye vision in terms of dividing the wide field of view into two zones the peripheral vision and the central vision. The peripheral vision included the entire field of view of the camera while the central vision was selected based on the path and lateral acceleration of the vehicle whether it is moving to the right, left or straight ahead. The results showed a great improvement in the computational time.